In unsupervised learning, you throw unlabelled data at a model and ask: 'What patterns do you see?'
The model clusters, compresses, or transforms the data to reveal structure.
**Use cases:**
- Customer segmentation (group users by behaviour)
- Anomaly detection (find unusual transactions)
- Dimensionality reduction (compress 1000-feature data to 2D for visualisation)
- Recommendation systems (find similar users/items)
Challenging because there's no 'right answer' — you need domain expertise to interpret what the model found.